Journal ofTesting and Evaluation
A. Hijazi1 and C. J. Kahler2
DOI: 10.1520/JTE20150437
Contribution of the ImagingSystem Components in theOverall Error of theTwo-Dimensional Digital ImageCorrelation Technique
VOL. 45 / NO. 2 / MARCH 2017
A. Hijazi1 and C. J. Kahler2
Contribution of the Imaging SystemComponents in the Overall Error of theTwo-Dimensional Digital ImageCorrelation Technique
Reference
Hijazi, A. and Kahler, C. J., “Contribution of the Imaging System Components in the Overall Error of the
Two-Dimensional Digital Image Correlation Technique,” Journal of Testing and Evaluation, Vol. 45, No. 2,
2017, pp. 1–16, doi:10.1520/JTE20150437. ISSN 0090-3973
ABSTRACT
Digital image correlation (DIC) is one of the most widely used non-invasive methods for
measuring full-field surface strains in a wide variety of applications. The DIC method has
been used by numerous researchers for measuring strains during the plastic range of
deformation where the strains are relatively large. The estimation of the amount of
background strain error in the measurements is of prime importance for determining the
applicability of this method for measuring small strains (such as the elastic strains in
metals, ceramics, bone samples, etc.). In this study, the strain errors in 2D-DIC
measurements associated with different types of imaging systems were investigated.
In-plane rigid-body-translation, experiments were used to estimate the overall amount of
error in DIC displacement and strain measurements. Different types of cameras having
different types of sensors and different spatial resolutions were used in the study. Also,
for the same type of camera, different types of lenses were used. Results show that the
DIC measurement accuracy depends on the magnitude of image displacement and that
different error estimation parameters can be used for quantifying the accuracy of the
measurements. Also, the effect of the lens on measurement accuracy is more pronounced
than that of the camera. Furthermore, imaging conditions such as image sharpness and
camera gain also affect the accuracy. Further still, the measurement accuracy was found
to be influenced by the direction of translation. The results indicate that measurement
error can be reduced by orienting the camera such that the major displacement direction
is parallel to the width direction of the image. The experimental approach used in this
study can be used for quantitatively assessing the quality of the different types of
Manuscript received October 14, 2015;
accepted for publication January 25,
2016; published online February 29,
2016.
1 Department of Mechanical Engineering,
The Hashemite Univ., Zarqa, 13115,
Jordan (Corresponding author),
e-mail: [email protected]
2 Institute of Fluid Mechanics and
Aerodynamics, Universitat der
Bundeswehr Munchen,
85577 Neubiberg, Germany.
Copyright VC 2016 by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959. 1
Journal of Testing and Evaluation
doi:10.1520/JTE20150437 / Vol. 45 / No. 2 / March 2017 / available online at www.astm.org
cameras and lenses and to determine their suitability for use in experimental techniques that depend on image analysis such
as DIC and particle image velocimetry (PIV).
Keywords
digital image correlation, error analysis, strain error, imaging system, camera, CCD, CMOS, lens distortion
Introduction
Nowadays, digital image correlation (DIC), sometimes referred
to as the digital speckle correlation method (DSCM), has
become one of the most widely used full-field optical methods
for motion and deformation measurements. The DIC method
was first introduced by Sutton et al. [1] in the early 1980s,
and during the past three decades it underwent continuous
modifications and significant improvements [2,3]. In its simpler
version, DIC is used for two-dimensional in-plane measure-
ments (2D-DIC) using a single camera. Also, photogrammetric
three-dimensional measurements (3D-DIC) [4] can be made
using two cameras in stereo configuration. Besides the good
measurement accuracy of the DIC method, it also offers other
attractive features which include a relatively simple experimen-
tal setup, simple or no specimen preparation, and low require-
ments for the measurement environment. All of that has made
the DIC method extremely popular among the experimental
mechanics community, and both 2D-DIC and 3D-DIC are
being increasingly used in a very wide range of applications
ranging from material science to mechanical, manufacturing,
biomedical, and structural engineering [5,6].
The basic idea of DIC is to compare two digital images,
acquired at different states (e.g., one before deformation and the
other one after), of a surface having a random speckle pattern
to determine the magnitude of displacement (or deformation)
between the two images. The comparison is done by dividing
the reference image into subsets of several pixels, then mathe-
matically matching those subsets with the other image (based
on intensity levels). By doing such, the new location of each
subset in the second image can be determined. From that, the
full-field deformation map can be obtained and the strain map
can then be easily determined.
Digital particle image velocimetry (DPIV, or usually just
referred to as PIV) [7–9] is a technique that is somehow similar
to DIC; however, it is used in experimental fluid mechanics for
obtaining the fluid velocity map within a region of interest.
Small seeding particles are introduced into the flow and the
kinematics of the fluid flow is estimated by tracking the motion
of these particles. The particles are chosen to have a density
such that they will be near naturally buoyant in the fluid in
order to faithfully follow the flow. The region of interest is illu-
minated using a thin planar light sheet (usually a laser is used
for illumination) such that the seeding particles within this light
sheet will be visible due to the light scattered off their surfaces.
The motion of the particles within the light sheet is captured
using a digital camera where typically two exposures with a
small inter-frame separation are recorded by the camera. By
comparing any two consecutive digital images (using techniques
quite similar to those used in DIC), the magnitude and direction
of motion of the seeding particles can be determined. Finally,
the magnitude of motion is divided by the inter-frame time sep-
aration between the two images such that the 2D velocity field
of the flow is obtained. Though DIC and PIV are used in totally
different applications, however, the two techniques are quite
similar in terms of the analyses performed on the captured
images. The first step in both techniques is basically the same
where the motion of each of the image subsets is determined,
though different correlation algorithms are typically used [3,10].
The difference between the two techniques comes in the second
step. In DIC, the motion map is used to calculate the strains,
whereas in PIV the motion map is divided by the inter-frame
time separation to calculate the velocity. Some of the commer-
cial software packages available nowadays are capable of
performing both DIC and PIV analyses.
In principle, 2D-DIC can be used for deformation measure-
ments under three conditions: the specimen has a planar sur-
face, it undergoes in-plane deformations, and the camera’s
optical axis is perpendicular to the specimen surface. If any of
these three conditions is not reasonably satisfied, the accuracy
of the measurements will be compromised. The measurement
accuracy of 2D-DIC depends on several factors, which include:
(a) the speckle pattern, (b) quality and perfection of the imaging
system (distortions, noise, resolution, etc.), and (c) the selection
of the correlation algorithm and parameters (subset and step
size, correlation and shape functions, sub-pixel algorithm, etc.)
[3,5]. Numerous studies have investigated the different sources
of error and tried to estimate the resulting errors and to suggest
remedies in some cases. Listings of the different studies can be
found in Refs. [3,5,11]. Similarly, there have been several studies
that addressed the accuracy of PIV measurements [10,12–14].
The majority of the studies aimed at investigating the accuracy
of the DIC or PIV measurements (especially for PIV) use simu-
lated digital images with artificial deformation patterns in order
to investigate the accuracy of the different correlation algo-
rithms and the effects of the different correlation parameters.
Journal of Testing and Evaluation2
In one of the key papers addressing the applications of DIC
in experimental mechanics, Chu et al. [15] used in-plane rigid-
body-translations to demonstrate the viability of the DIC
method for actual measurements. When a body undergoes a
rigid-body-translation, the measured strains should theoreti-
cally be zero. Thus, any obtained strain readings will actually
reflect an error in the measurement and the magnitude of
the obtained strains is simply the magnitude of error. In a
rigid-body-translation experiment, the camera is directed per-
pendicularly to a surface having a random black and while
speckle pattern, and an image is captured while the surface is at
its initial position, then the surface is rigidly translated in-plane
by a small amount and another image is captured. This simple
experiment remains to be one of the most widely used experi-
ments for estimating the magnitude of background strain error
expected in DIC strain measurements. Several other researchers
have also used in-plane rigid-body-translation experiments to
estimate the expected background error in DIC measurements
under several experimental conditions. Haddadi et al. [16] did
an experimental investigation in which they used rigid-body-
translations to investigate different sources of error in 2D-DIC
measurements and to estimate these errors. In that study, they
estimated the strain errors associated with the magnitude of
in-plane translation, subset and step sizes, in-plane rotation,
speckle pattern, out-of-plane displacement, lightening, and test-
ing environment. Sutton et al. [17] studied the effects of out-of-
plane displacements and rotations (that may occur during the
loading) both theoretically and experimentally using rigid-
body-translations. Their results show that out-of-plane transla-
tion could lead to significant error in 2D-DIC strain measure-
ments, especially for short camera-to-object stand-off distance.
They also showed that the use of telecentric lenses minimizes
the error to a manageable level. Hijazi et al. [18] also used rigid-
body-translations as well as theoretical analysis to study the
influence of camera non-perpendicularity on the measurement
accuracy of 2D-DIC. Their results showed that small amounts
of misalignments could result in relatively large errors in strain
measurements, especially if the camera-to-object stand-off dis-
tance is short. They also showed that the rigid-body-translation
experiments can be used for verifying the perpendicularity of
the camera with respect to the surface being observed.
Researchers have also studied the influence of the lens dis-
tortions on 2D-DIC measurement accuracy for both micro-
scopic and macroscopic levels [19–22]. In fact, the effect of the
lens is more obvious for 2D-DIC than it is for 3D-DIC. This
is simply due to the fact that 3D-DIC measurements involve a
rigorous pre-measurement calibration procedure, which can
correct lens distortions to a reasonable extent. However,
2D-DIC, on the other hand, is generally used without any pre-
measurement calibration where this is regarded as one of the
attractive features for this method. Pan et al. [21] investigated
the 2D-DIC strain measurement error due to lens distortion
both theoretically and experimentally. They proposed a simple
first order lens distortion correction method to correct for radial
distortion and they applied it to images recorded using a tele-
centric lens during in-plane rigid-body-translation experiments.
Similarly, Lava et al. [22] also studied the impact of lens distor-
tions on 2D-DIC strain measurements accuracy using in-plane
rigid-body-translation experiments. They developed a correc-
tion method that corrects for radial and tangential lens distor-
tions and they applied their calibration procedure for three
lenses having different focal lengths (12, 23, and 50mm). Rue
et al. [23] studied the influence of the imaging system (camera
and lens) resolution on 2D-DIC measurements. They used syn-
thetic and experimental images with different resolutions and
concluded that a careful choice for the camera/lens combination
should be made for DIC measurements where one of the two
components can be resolution limiting. Barranger et al. [24]
investigated the effect of the camera dynamic range (i.e., gray
scale bit depth) on DIC measurement accuracy using in-plane
rigid-body-translation experiments. They compared three dif-
ferent cameras with 8-bit, 10-bit, and 12-bit dynamic range and
they observed that the 8-bit camera gives higher accuracy than
the ones with higher dynamic range. Tiwari et al. [25] per-
formed an assessment for the applicability of using high-speed
cameras (frame rates higher than 1000 fps) for DIC measure-
ments. They compared two types of high-speed cameras, an
ultra-high-speed multi-channel intensified CCD camera and a
high-speed CMOS camera. Their results showed that the CMOS
camera gives slightly better accuracy for DIC measurements
and that both cameras can yield reasonably accurate strain mea-
surement after applying a calibration procedure to remove
image distortions.
From the above literature survey it is evident that there is a
lack of studies that discretely investigate and quantify the effect
of the imaging system components (both cameras and lenses)
on 2D-DIC measurement accuracy. Comparing different types/
classes of cameras and lenses that are widely used in DIC and
identifying the contributions of the cameras and lenses in the
measurements error is of a particular interest for many DIC
users. In the work presented in this study, the strain errors
in 2D-DIC measurements associated with different types of
imaging systems were quantified and compared. A rigorous
experimental procedure that uses in-plane rigid-body-
translation experiments was used to determine the overall
amount of error in DIC displacement and strain measurements
in order to compare the different cases that were investigated
here. Different types of cameras having different types of sen-
sors were used in the study. Also, for the same type of camera,
different types of lenses were used. In addition, different mea-
surement error estimation parameters that are typically used for
DIC accuracy measurement were compared. Furthermore, the
effect of the direction of the in-plane translation on measure-
ment error was also investigated.
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 3
Digital Imaging Devices
An image capturing device consists of an imaging optic (i.e.,
lens) which collects the light emanating from a target and
forms an image of that target on a light sensitive medium
(electronic image sensor, photographic film, etc.). An elec-
tronic image sensor consists of a matrix of capacitor-like
storage elements, known as pixels, formed on an oxide cov-
ered silicon substrate. This type of sensor, which is known as
the Metal Oxide Semiconductor (MOS) sensor, relies on the
photoelectric property of silicon to convert the incident light
to electrical charge. As an optical image is projected on the
imaging sensor, which is usually referred to as the focal
plane array (FPA), the photons reaching each pixel generate
an electrical charge, usually electrons, the magnitude of
which is proportional to the local intensity of light on that
pixel. After the sensor has been exposed to light for a period
of time (the integration or exposure time), a pattern of
charges is collected in the pixels (i.e., a frame is captured).
The pattern of charges is then readout to a storage device,
freeing the sensor to capture another image. The two most
widely recognized types of MOS sensors are the complemen-
tary metal oxide semiconductor (CMOS) and the charge-
coupled device (CCD). Both the CMOS and CCD sensors
were invented around the same time; however, due to the
more complicated design of CMOS sensors, the CCD tech-
nology developed much faster and CCD sensors became
more dominant. The technological advances during the last
two decades made the manufacturing of CMOS sensors more
economical and thus they are increasingly being used as an
alternative to CCD sensors in many scientific, industrial, and
consumer cameras. The basic difference between CCD and
CMOS sensors is in the way an image (i.e., the pattern of
charges collected in the pixels) is transferred out of the sen-
sor after it has been captured [26]. There are three common
types of CCD sensors which are the “full frame,” the “frame
transfer,” and the “interline” CCDs [27]. The main difference
between these three types is in the layout of the sensor where
each of the three different designs has its advantages and dis-
advantages. A thorough review of the different types of elec-
tronic imaging sensors can be found in Ref. [26] and a
comparison of the performance of different types of imaging
sensors can be found in Ref. [28].
Unlike compact digital cameras used in regular photogra-
phy, which have an electronic sensor and a lens that are inte-
grated into a single unit, imaging systems used in scientific
applications usually come as two separate units: the camera
body and the lens (similar to professional photography cam-
eras). The cost of cameras used in scientific applications vary
significantly according to the type of sensor, its resolution, sig-
nal to noise level, readout speed, frame rate, etc. The same also
applies to lenses according to the quality of the optics and the
amount of optical distortions. Consequently, the cost of imaging
systems used in scientific applications could vary from less than
a thousand dollars to more than $100,000 for some of the high
speed systems.
Experimental Procedures
In this study, in-plane rigid-body-translation experiments were
used to investigate the strain errors in 2D-DIC measurements
associated with different types of imaging systems. Different
classes of cameras having different types of sensors, resolutions,
and dynamic range (bit depth) were used in the study. Also, for
the same type of camera, different types of lenses were used in
order to investigate the effect of the lens on strain error. The
different cameras and lenses that were used in this investigation
along with the main specifications for each are listed in Tables 1
and 2, respectively.
As can be seen from Table 1, four different cameras were
used in this investigation. The SensiCam camera is one of the
high end scientific cameras; it features a cooled FPA which pro-
vides high stability and quantum efficiency. Also, this camera is
capable of operating in a special mode known as “dual-frame
mode,” where it can capture a pair of images with very short
inter-frame time [26], and thus it is commonly used in PIV
applications. The Genie camera is one of the moderate cost
CCD industrial cameras, whereas the Photon Focus camera has
a CMOS sensor with higher dynamic range, and thus its price is
higher. The Canon camera is one of the typical high resolution
SLR digital cameras used in professional photography. Though
it will be elaborated later in the results and discussion section, it
is also worth mentioning here that the Canon color-SLR digital
camera is in fact not comparable to the other cameras that were
used here, and even it was not compared on one-to-one bases in
terms of the lens (as can be seen in Table 3). However, the
TABLE 1 The main specifications of the different cameras used in the experiments.
Camera PCO / SensiCam QE DALSA / Genie-M1410 Photon Focus / MV1-D1312-80 Canon EOS 450D
Sensor Monochrome 2/3 in. interline CCD Monochrome 2/3 in. interline CCD Monochrome 1 in. CMOS Color 11=4 in. CMOS
Resolution 1376� 1040 1360� 1024 1312� 1082 4272� 2848
Dynamic range 12-bit 8-bit 12-bit 24-bit (color)
Frame rate 10 fps 22 fps 55 fps 3.5 fps
Lens mount C-mount C-mount C-mount EF-mount
Journal of Testing and Evaluation4
Canon camera was included in this investigation simply because
similar cameras are used by some DIC users, in addition to the
fact that some DIC software companies commercialize similar
cameras as an alternative for applications where DIC measure-
ments need to be done in the field.
Table 2 lists the four different types of lenses that were used
in this investigation. The Zeiss lens is a high quality lens with
high aperture (i.e., a fast lens) and very low optical and chro-
matic aberrations. The Pentax lens is a small size, good quality
lens designed for machine vision applications and it comes at a
moderate price. The Nikon zoom lens is an old model lens that
was designed for use with old-style film SLR cameras. The
Canon zoom lens is the default lens that is supplied with the
Canon EOS camera, and it can be operated in the automatic or
manual focus modes.
In order to investigate the influence of the different types of
cameras and lenses on DIC measurements accuracy, different
combinations of the cameras and lenses were used in the experi-
ments as listed in Table 3. As can be seen in the table, six differ-
ent camera/lens combinations were used in the experiments.
The SensiCam/Zeiss, Genie/Zeiss, and Photon Focus/Zeiss com-
binations were used to study the effect of the camera, whereas
the Genie/Zeiss, Genie/Pentax, and Genie/Nikon were used to
study the effect of the lens. An F-mount to C-mount adaptor
was used to fit the Zeiss and Nikon lenses on the cameras. The
Canon camera, on the other hand, was only used with its own
lens mainly because the other lenses do not fit this camera, as
it has a large size FPA. Also, the Canon camera is basically dif-
ferent from the other cameras due to the fact that it is a color
camera and it is not very common for this type of camera to be
used for DIC or PIV applications.
Besides the experiments performed using the different cam-
era/lens combinations, two additional experiments were per-
formed using the Genie/Zeiss combination, but under different
imaging conditions. One of the two additional experiments was
performed while the lens was slightly defocused to produce
slightly blurred images, whereas in the other one, the camera
gain was set to a high value of 11 dB (the default setting of the
camera is zero gain). In the experiment with high gain setting,
the intensity of illumination was reduced to compensate for the
increased gain such that the average image intensity level was
comparable to the other experiments.
In the experiments conducted in this study, close attention
was paid to ensure that the same experimental conditions, set-
tings, and procedures were used for all experiments in order to
obtain a one-to-one comparison between the different cameras/
lenses that were investigated. A picture of the experimental
setup is shown in Fig. 1. In each of the experiments, the camera
was mounted on a multi-axis translating/rotating stage to allow
adjusting the position of the camera with respect to the target
and the same multi-axis stage was used for performing the
rigid-body-translation experiments. To avoid any confusion, it
might be worth mentioning here that the camera was translated
during the experiments rather than translating the target and
that basically gives the exact same outcome (that was done sim-
ply because the same setup was also used for other experiments
that are out of the scope of this paper). A printed random
speckle pattern having black dots on a white background was
attached to a flat plate. The average diameter for the dots of the
speckle pattern was about 0.5mm and the average dot center-
to-center spacing was about 1.2mm. The target was illuminated
using an adjustable intensity illumination source coupled with a
fiber-optic delivery cable. The distance between the front end of
the lens and the target surface (i.e., the working distance) was
set to be between 750 and 800mm according to camera/lens
combination being used. The camera-target working distance or
the zoom level (for zoom lenses) was adjusted such that the
field-of-view observed by the camera is 100mm wide for all
camera/lens combination that were used. Since the field-of-view
width was fixed for all experiments, the image scale-factor for
TABLE 2 The main specifications of the different lenses used in the experiments.
Lens ZeissMakro-Planar - ZF Pentax C5028-M Nikon Series-E Zoom Lens Canon EF-S Zoom Lens
Focal length 50mm 50mm 75 – 150mm 18 – 55mm
Aperture f/2 - 22 f/2.8 -22 f/3.5 - 32 f/3.5 - 5.6
Mount F-mount C-mount F-mount EF-mount
TABLE 3 The different camera/lens combinations used in the experiments along with the abbreviations used for identifying each of the cameras and
lenses.
Camera SensiCam (SC) Genie (Gn) Photon Focus (PhF) Canon (Cn)Lens
Zeiss (Zs) � � �Pentax (Pn) �Nikon zoom (Nk-z) �Canon zoom (Cn-z) �
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 5
the different cameras varied according to the camera resolution.
The scale-factors for the different cameras used here were: 13.8
pixels/mm for SensiCam, 13.6 pixels/mm for Genie, 13.1 pixels/
mm for Photon-Focus, and 42.7 pixels/mm for Canon. Also, for
all camera/lens combinations, the camera position was initially
adjusted to capture the exact same region of the speckle pattern.
In each of the experiments conducted using the different
camera/lens combinations, great care was taken to ensure the
perpendicularity of the camera with respect to the target surface
to prevent any measureable errors that might be caused by cam-
era non-perpendicularity from hindering the results of this
study. An electronic level with laser pointer was used for orient-
ing the cameras perpendicular to the target surface. The camera
perpendicularity was then verified using rigid-body-translation
experiments based on the procedure presented in Ref. [18]
where the perpendicularity error (if any is present) was deter-
mined to be less than 0.5�. Also, care was taken to obtain similar
image brightness levels (an average image brightness of about
150 on 8-bit grayscale) regardless of the camera/lens combina-
tion being used, except for the Canon camera, which was run in
its “Auto” mode. The image brightness was controlled by
adjusting illumination intensity while the aperture of the lenses
was set at an intermediate value (around f/11). Furthermore, all
the cameras that were used in this study were allowed to run for
1 h, in order to let the FPA temperature to reach steady state
(such that the image noise level is stabilized), before starting
to capture images for the experiments. This is particularly
important for the SensiCam camera since it has a cooled FPA.
It should also be noted here that some of the cameras used
here have dynamic ranges higher than 8-bit, as can be seen
from Table 1. However, in order to eliminate any effect that the
dynamic range might have on the results, all the images used
for the experiments were recorded at 8-bit. For the color images
captured by the Canon camera, image processing software was
used for converting the images into 8-bit grayscale images.
The exact same rigid-body-translation experiment was
repeated for all camera/lens combinations. Starting from the
initial position, the camera was translated in two steps along the
x-direction, then in two steps along the y-direction and two
duplicate images were captured at each position. Each of the
two translation steps, in both the x and y directions, was equal
to 8 mm (8 % of the field-of-view width). As a result, images
were captured while the camera was at five different positions; a
reference position (i.e., the position after the first two transla-
tion steps in the x-direction), two translation steps in the
x-direction, and two translation steps in the y-direction. Fig. 2
illustrates the positions of series of images captured during the
experiments relative to the fixed target. The rectangular frames
shown in the figure represent the boundaries of the image in
each of the five different positions, where the solid line identifies
the reference position and the dashed and dotted lines identify
the two translated positions along the x and y directions. The
figure also shows the square region of interest used for DIC
analyses (identified using dashed line).
FIG. 1
The experimental setup.
Journal of Testing and Evaluation6
Analyses
DIC ANALYSES
The analyses were performed using a 2D-DIC software called
MatchID-2D [29]. In order to use the exact same region of the
image in all DIC analyses, a square region of interest corre-
sponding to a size of about 59� 59mm of the speckle pattern
was located in the upper central region in the reference image,
as illustrated in Fig. 2. However, the size of this region of inter-
est in pixels was different according to the resolution of the
camera being used. Also, since the cameras have different reso-
lutions, different DIC parameters in terms of subset size and
step size were used in order to have a direct one-to-one compar-
ison between the different cameras. Table 4 lists the size of the
region of interest, subset size, and step size (in pixels) used for
DIC analyses for the images captured by each of the four differ-
ent cameras. From the table, it can be seen that the step was
taken to be about half of the subset size in all cases in order to
get 50 % overlap between adjacent subsets. It also should be
noticed that though different values are used for the different
cameras (as seen in the table), they actually correspond to about
the same physical size on the speckle pattern (59� 59mm
region of interest, 2.25� 2.25mm subset size, and 1.1mm step
size) since the image scale-factors are different. The Normalized
Cross-Correlation algorithm was used for the DIC analyses of
all the different experiments and no image pre-filtering was per-
formed on the images. After obtaining the displacement maps
in both x and y directions (i.e., u and v), the Green-Lagrange
strains were calculated using a strain window size of 7� 7
points for all cases (that corresponds to a physical size of about
6.6� 6.6mm for the strain window). No further smoothing was
performed on the obtained strain maps.
As mentioned earlier, two duplicate images were recorded
by the camera while it was at each of the different positions dur-
ing the rigid-body translation experiments. These duplicated
images were correlated with each other where one of the images
was taken as the reference image, while the other was taken as
the deformed image. Since the two images were captured at the
same position, the correlation should give zero displacement at
all points in the x and y directions. However, due to the noise in
the digital images which cause some random fluctuation in the
image intensity values, the DIC results show very small random
sub-pixel displacements at all points. The displacements
obtained from such correlation represent the baseline error in
DIC analysis. Also, such correlation is useful in verifying
whether the correlation parameters being used are appropriate
or not. For each camera/lens combination, the correlations were
performed between each of the five pairs of duplicated images,
FIG. 2
The positions of the series of images captured
during the experiments.
TABLE 4 The DIC size parameters for the different cameras used in the experiments.
Camera SensiCam Genie Photon Focus Canon
Region of interest (pixel) 800� 800 800� 800 780� 780 2520� 2520
Subset size (pixel) 31� 31 31� 31 29� 29 97� 97
Step size (pixel) 15 15 14 48
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 7
which correspond to the five different positions during the
rigid-body translation experiment, and the results of the five
correlations were averaged to get a more reliable estimate of the
baseline error.
For each group of rigid-body-translation experiments
corresponding to a different camera/lens combination, the ref-
erence position image was correlated with the images corre-
sponding to each of the two translation steps in each of the two
directions. Since two duplicated images were captured at each
position, a total of four duplicated correlations were performed
for each translation step (e.g., position 1-1 & position 2-1, posi-
tion 1-1 & position 2-2, position 1-2 & position 2-1, and
position1-2 & position 2-2). The results of the four replicates
were averaged in order to eliminate any variation in the results
that might be caused by the instability of the image intensity
values (though such differences were found to be very small).
ERROR ANALYSES
As mentioned earlier, the measurement accuracy of 2D-DIC
depends on several factors, and numerous studies have investi-
gated the influence of the different sources of error on the accu-
racy of the results. However, there is still a lack of a common or
standard procedure for evaluating the accuracy of DIC meas-
urements. Patterson et al. [30] proposed and presented standar-
dized test material and procedure for the evaluation of optical
techniques and systems used for full-field strain measurements.
They demonstrated the use of their proposed approach for DIC
and electronic speckle pattern interferometry (ESPI); however,
that approach is still not widely accepted among DIC users.
Hoult et al. [31] compared DIC strain measurements to the
strains measured using strain gauges during tensile testing.
They averaged the DIC strain values within a region of interest
and simply compared the averaged strain value to the strain
measured using strain gages to evaluate the accuracy of DIC
strain measurements. However, this simple approach does not
evaluate the accuracy of DIC strain maps since it is based on an
averaged value for DIC strain.
In general, the vast majority of studies related to DIC accu-
racy or error analysis can be grouped into two broad categories
according to their procedure. In the first category, synthetic
images are used where an image of a speckle pattern is numeri-
cally modified. A known amount of translation, rotation, homo-
geneous or heterogeneous deformation, is applied to the image;
then this new modified image is correlated to the reference (i.e.,
initial) image to determine the deformation. The results are
then compared to the applied translations or deformations in
order to assess their accuracy [11]. These types of studied are
useful for assessing the relative accuracy of the different correla-
tion algorithms as well as evaluating the effect of different
parameters (e.g., correlation parameters, speckle pattern, image
intensity, image contrast, etc.) on the accuracy of the results.
However, the effect of the imaging system and its optical
components on DIC accuracy is not accounted for in the error
estimates obtained by such studies. The use of such synthetic
images is very common for evaluating the accuracy of the differ-
ent correlation algorithms in PIV analyses [10].
The second category of studies basically uses in-plane rigid-
body-translation (or rotation) experiments to investigate the
accuracy of DIC measurements [15–18,21–24]. In rigid-body-
translation experiments, the target surface can be translated
with a known magnitude and direction, but, most importantly,
all points on the surface are translated by the same amount;
thus, the strains are zero. The accuracy of DIC measurements
can be assessed based on the displacement or strain results.
When the error assessment is based on displacement, usually,
the standard deviation of the obtained displacement values of
all points within the region of interest is calculated. Theoreti-
cally, all points have the same displacement during rigid-body-
translation and thus the standard deviation should be zero.
Therefore, the value of the displacement standard deviation,
usually reported in pixel units, is considered to represent the
magnitude of error in the displacement measurements. This
kind of displacement error estimation is usually used in PIV
analyses [12,13]. On the other hand, when the error assessment
is based on strains, any strains obtained from the DIC analysis
reflect an error in the results since the strain should in fact be
zero. The “mean” strain value (ð1=NÞP
e) for all points within
the region of interest can be calculated; however, this value will
be meaningful only when the strains obtained at all points are
positive or negative. This will be the case if the camera was not
perpendicular to the surface or if an out-of-plane displacement
occurred during the experiment [18]. Instead, the mean of the
strain absolute values (ð1=NÞP
ej j) is calculated where it can be
considered as a measure of the magnitude of error in DIC strain
measurements [16,18]. In addition, the standard deviation of
the strain values� ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP
e� �eð Þ2=Nq �
is also considered by many
researchers as a measure of the magnitude of error in DIC strain
measurements [11]. Some other researchers report the maxi-
mum strain value as a representation of the magnitude of strain
error [22]; however, such an approach is rarely used since it
exaggerates the magnitude of error.
Results and Discussion
THE DIFFERENT DIC ERROR ESTIMATION PARAMETERS
As mentioned in the previous section, the accuracy of DIC
analyses can be assessed based on the displacements or strains
obtained during rigid-body translation experiments. Three dif-
ferent parameters are commonly used for estimating the magni-
tude of error in DIC analyses and these parameters are: (1) the
standard deviation of the obtained displacements, (2) the mean
of the absolute values of the obtained strains, and (3) the stand-
ard deviation of the obtained strains. The standard deviation of
Journal of Testing and Evaluation8
the obtained displacement field is commonly used for reporting
the accuracy in PIV analyses since no subsequent strain calcula-
tions are done. The accuracy (or magnitude of error) in such a
case is usually given in pixel units, or it can be converted to dis-
placement units (mm or so) knowing the scale-factor of the
digital images. For DIC analyses, it is more common to report
the accuracy as the magnitude of error in the strain measure-
ments (the mean of absolute values or the standard deviation).
The three different error estimation parameters were calculated
for all the experiments performed in this study. Fig. 3 shows a
comparison of the three different DIC error estimation parame-
ters for one of the camera/lens combinations (SC/Zs) during
rigid-body-translation along the x-axis. As can be seen in the
figure, the three different error estimation parameters show a
somewhat similar increasing trend as the magnitude of the
rigid-body translation increases. This trend of increasing error
as the magnitude of displacement increases can be seen in all
the different experiments performed in this study and it is also
consistent with the results reported in literature [5,11,15,16,18].
Comparing the mean of absolute values and the standard devia-
tion of the obtained strains, it can be seen that the standard
deviation has a consistently higher magnitude. This difference is
quite understandable knowing that the errors are randomly dis-
tributed, and they usually follow a normal distribution [18,22].
Thus, the standard deviation value is bigger than about 68 % of
the data points, whereas the mean of absolute values is practi-
cally bigger than about 50 % of the data points. Thus, since the
standard deviation gives a more conservative estimate of the
strain error (since it has a higher value), it will be used for rep-
resenting the magnitude of DIC strain measurement error for
comparing the different camera/lens combinations.
Furthermore, it can also be seen from Fig. 3 that the dis-
placement error increases at a much faster rate compared to the
strain error, as the magnitude of translation increases. By
inspecting the values in the figure, it can be seen that the error
in the measured displacements increases from about 0.005 pixel
when the magnitude of translation is zero to about 0.039 and
0.076 pixels for 8 and 16mm of rigid-body-translation, respec-
tively. This observation calls for caution when dealing with PIV
or DIC displacement error values reported in literature where
any value must be associated with its corresponding magnitude
of translation. The fast increasing trend of the displacement
error, as compared to the strain error, can be attributed to the
fact that the strain is calculated using a relatively large strain
window (7� 7 points strain window size was used in this study,
whereas the smallest possible window size setting is 3� 3
points). As the strain calculation window size gets larger, more
displacement data points are used for calculating each strain
data point, and this results in smoothing out the random error
in the measured displacements and thus the magnitude of the
measured strain error is reduced. However, the use of large
strain windows is common in DIC analyses where some
researchers report using a strain window as large as 21� 21
points [21]. In general, the use of large strain window sizes
reduces the strain error, but at the same time, it reduces the
effective spatial resolution of the strain map and thus can
hinder any high strain gradients that might be present. Pan
et al. [5] recommend using large strain calculation windows for
measuring homogeneous deformation. On the other hand, for
inhomogeneous deformation, they generally recommend
smaller strain calculation windows such that a balance can be
obtained between accuracy and smoothing.
Fig. 4 provides another insight regarding PIV or DIC
displacement error. The figure shows a comparison of the dis-
placement errors (in pixels) for four different camera/lens com-
binations. It can be seen in the figure that the displacement
error for the Canon camera is much higher than it is for all the
FIG. 3 Comparison of the three DIC error estimation parameters (mean of
absolute strain values, strain standard deviation, and displacement
standard deviation) for rigid-body translation along x-axis (SC/Zs).
FIG. 4 Comparison of the DIC displacement error for four different camera/
lens combinations.
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 9
other cameras. In fact, there are reasons for the higher DIC
measurements error associated with the Canon camera, as will
be discussed in a later section. However, the large difference
seen in Fig. 4 is mainly attributed to the fact that this camera
has a much higher image scale-factor than the other cameras
(due to its much higher digital resolution). For instance, for
16mm displacement, the displacement error for the Photon-
Focus camera is 0.18 pixels, while for the Canon camera, the
error goes to 1.18 pixels (about six folds the magnitude). How-
ever, since the scale-factor for the two cameras is quite different
(13.1 pixels/mm for Photon-Focus, and 42.7 pixels/mm for
Canon), if we convert these displacement error values and rep-
resent them in millimeters, the difference goes down and the
displacement error for the Photon-Focus becomes 0.013mm,
whereas for the Canon it becomes 0.028mm (about two folds
the magnitude, only). This observation brings to attention that
reporting the displacement error in pixel units can be mislead-
ing if it is not associated with the scale-factor of the digital
images. In fact, reporting the displacement error in pixels as an
independent measure of accuracy (i.e., without reporting the
associated scale-factor) can lead to a false impression that
higher measurement accuracy can simply be attained by using
higher resolution cameras.
CAMERA EFFECT ON DIC STRAIN ERROR
A comparison showing the effect of the camera on DIC strain
measurement error is presented in Fig. 5. The figure shows the
strain error (standard deviation) for four different cameras. For
the first three cameras (SensiCam, Genie, and Photon-Focus),
the same lens was used in the experiments such that a direct
one-to-one comparison can be made between the cameras.
However, the fourth camera (Canon) was used with the zoom
lens supplied with it since the other lenses do not fit this cam-
era. Moreover, it should be noted that the Canon camera used
here is a regular photography digital-SLR color camera and it is
not intended to be used for scientific applications. Nevertheless,
some DIC software companies commercialize similar cameras
as an alternative for field measurements. Also, some researchers
report using similar SLR digital cameras in DIC measurements
[16,31,32]. Thus, the Canon camera was used in this study in
order to compare its performance in DIC measurements with
other cameras that are typically used in DIC or PIV applica-
tions. As can be seen in the figure, for all of the four cameras,
the magnitude of strain error increases as the magnitude of dis-
placement increases. For the Canon camera, though the strain
error at zero displacement is relatively small, it increases at a
faster rate as the displacement increases. The high magnitude of
strain error for the Canon camera relative to the other cameras
is most likely due to two reasons. Firstly, and most importantly,
comes the fact that this is a color camera; secondly, that a zoom
lens is used with this camera. Color cameras are equipped with
imaging sensors that are covered with a mosaic pattern of red,
green, and blue filters. As the target translates, points on the tar-
get that were initially imaged through one of the color filters get
imaged through a different color filter and so on, causing varia-
tion in the image intensity at the pixel level between successive
images. Regarding the zoom lens used with the Canon camera
in general, it is well known that zoom lenses produce lower
quality images than fixed focal length lenses. This is because
optical aberrations cannot be effectively corrected at the entire
focal length range of the zoom lens. However, it will be seen
from the next section that the use of zoom lens will not result in
very significant increase in error such as that seen with the
Canon camera. In summary, though the Canon camera digital
resolution is much higher than the other cameras used in this
study, the fact that it has a color sensor and the zoom lens used
with it, evade any benefit of the height digital resolution, if any.
In fact, Reu et al. [23] report that there is no benefit for using
high resolution cameras in DIC measurements if the lens reso-
lution is lower than that of the camera.
As mentioned earlier, the magnitude of strain error at zero
displacement represents the baseline error value for any cam-
era/lens combination. The figure shows that strain error, at zero
translation, is lower for the Photon-Focus and Canon cameras
than that of the other two cameras. This difference can be
attributed to the fact that the Photon-Focus and Canon cameras
have CMOS sensor, whereas the other two cameras have CCD
sensor. As mentioned previously, the CMOS sensors represent
the latest technology in imaging sensors and, in general, they
provide better stability in terms of the image intensity levels as
compared to CCD sensors. A study done by Hain et al. [28],
which compared CCD and CMOS sensors, showed that CMOS
sensors have better signal-to-noise ratio (SNR) than CCD sen-
sors. Thus, this lower strain error level at zero displacement is
simply due to the higher stability in image intensity levels
resulting from the height SNR of the CMOS sensors.
FIG. 5 Comparison of the camera effect on DIC strain error.
Journal of Testing and Evaluation10
Comparing the strain errors for the SensiCam and Genie
cameras, it can be seen that almost no difference can be seen
between the results of the two cameras. These results suggest
that the relatively low cost industrial machine vision cameras
may perform as good as the more expensive specialized scien-
tific cameras such as the SensiCam (however, it should be kept
in mind that the SensiCam is capable of capturing image pairs
with a inter-frame time of a few microseconds, which is a capa-
bility not available in the industrial machine vision cameras).
Also, it can be seen from the figure that though the strain error
for the Photon-Focus camera is clearly lower than that of the
SensiCam and Genie at zero displacement, it increases and
becomes slightly higher than that of the two cameras when the
target is translated. As discussed previously, the low strain error
at zero displacement is due to the stability of the image intensity
levels for the CMOS sensors. However, CMOS sensors are
known to have lower fill factor compared to CCD sensors due
to the presence of the digitization circuitry for each pixel on the
sensor itself [26]. Thus, when translation occurs, the low fill fac-
tor of the CMOS sensor comes into effect and, apparently, it
evades the advantage of the higher stability of the image inten-
sity levels.
Finally, the results presented in Fig. 5 show that the level of
the strain error obtained with three of the cameras tested here is
about 0.0003 (i.e., 300 micro-strains) for 8mm displacement
(which represents 8 % of the field of view width, as mentioned
earlier). This level of strain error seems to be very low when
compared to plastic strains in ductile materials. However, tradi-
tional strain measuring devices, such as strain gauges for
instance, still provide strain measurements with higher accuracy
than DIC. A comparison done by Patterson et al. [30] showed
that electronic speckle pattern interferometry (ESPI) and strain
gauges give higher accuracy than DIC for the measurement of
elastic strains. Hild and Roux [33] demonstrated that DIC can
be employed for the identification of elastic properties of low
stiffness materials using a Brazilian disk made of polycarbonate
polymer. In general, the DIC method is a powerful and widely
accepted method for measuring strains in the plastic range of
deformation (or in the elastic range for low stiffness materials
such as polymers) where the strains are relatively large, but, it
found very limited success for measuring small strains such as
the elastic strains in stiff materials (e.g., metals and ceramics).
For instance, if we take carbon steel as an example, the maxi-
mum elastic strain for most carbon steels is in the range
0.002–0.004. The results presented here imply that though the
magnitude of the DIC strain error is relatively small (about 300
micro-strains), but still it is not considered that small when
compared to the values of elastic strains of stiff materials (such
as carbon steel for example). Based on that, it is reasonable to
believe that DIC method is not a good choice for accurately
measuring elastic strains in stiff materials, especially when high
strain gradients are present. It might be worth mentioning here
that though smaller strain error estimates might sometimes be
found in literature, that does not necessarily reflect higher accu-
racy. It should be kept in mind that it is possible to further
reduce the estimated strain error by using larger strain window
size (a 7� 7 points strain window was used in this study, as
mentioned earlier) and/or employing a filter to smooth the
strain map. However, doing such will result in reducing the
effective resolution of the 2D strain map.
The results presented here for comparing the different cam-
eras shows that the rigid-body-translation experiments and the
approach followed in this paper can be used as a simple and
direct method for evaluating and comparing the metrological
performance of cameras and in particular their suitability for
use in full-field deformation measurement [28,34].
LENS EFFECT ON DIC STRAIN ERROR
Fig. 6 shows a comparison of the DIC strain error values
obtained using three different lenses (Zeiss, Pentax, and Nikon
zoom) with the same camera (Genie). From the figure it can be
clearly seen that there is an appreciable difference between the
strain error values obtained using the three different lenses. In
fact, the difference in strain errors seen here is more pro-
nounced than that seen previously in Fig. 5 (except for the
Canon camera), which indicates that the effect of the lens on
DIC strain measurement accuracy can be more significant than
the effect of the camera. The figure shows that the Zeiss lens
gives the highest accuracy rather than the Pentax lens then the
Nikon zoom lens. This trend is actually not surprising where it
is consistent with the known quality of these lenses (it is even
consistent with the price tags of these lenses). As in the previous
figures, the usual trend of increasing strain error as the displace-
ment increases can also be in this figure. The reason behind the
difference in the magnitude of strain measurement error
obtained using the different lenses can simply be attributed to
the presence and severity of optical aberrations in these lenses.
FIG. 6 Comparison of the lens effect on DIC strain error.
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 11
Different types of optical aberrations such as field curvature,
coma, distortion, astigmatism, spherical, etc., can be present in
lenses [35]. Lens manufacturers design their lenses in order to
correct these aberrations, but the effectiveness and accuracy of
these corrections vary between the different types of lenses.
Fig. 7(a) and 7(b) show the DIC horizontal displacement “U”
maps obtained from the images captured using the Zeiss and
Pentax lenses at zero, 8, and 16mm rigid-body-translations.
From the figure, it can be seen that the “U” displacement error
looks random at zero translation, while there is a clear and dis-
tinct displacement error pattern associated with each of the two
lenses that can be seen at both 8 and 16mm translations. The
displacement error patterns seen in the figure reflect the optical
distortion in the images formed by each of the two lenses. How-
ever, at zero translation, the lens distortion cannot be captured
by DIC analysis since the image is correlated with a reference
image captured while the camera is imaging the same position
(two duplicate images). But nevertheless, by referring again to
Fig. 6, it can be seen that the effect of the lens on DIC strain
error is also present at zero translation, which is a bit surprising
but still explainable. The lens effect on DIC strain error seen at
zero translation is basically related to the difference in definition
(i.e., sharpness) of the optical images formed by the different
lenses. Such difference cannot be recognized by the naked eye
in many cases; however, its effect is captured by the DIC analy-
sis. Higher definition images of the black and white speckle
pattern will show steeper change in the intensity levels of the
digital images between the black and white regions. This steep
change in intensity levels makes the matching of image subsets
more accurate and thus improves the displacement and strain
measurements accuracy.
By referring again to Fig. 7 and inspecting the shape of dis-
placement error patterns seen at 8 and 16mm translations for
each of the two lenses, it can be seen that the shapes of the error
patterns for the two lenses are quite different. The difference in
the error pattern shapes indicates that different types of optical
FIG. 7
DIC horizontal displacement “U” maps at Dx¼0, D x¼ 8 mm, D x¼ 16 mm, (a) Zeiss lens,
(b) Pentax lens.
Journal of Testing and Evaluation12
distortions are present in the two lenses. As mentioned earlier,
Pan et al. [21] and Lava et al. [22] proposed mathematical mod-
els and procedures for correcting for radial, radial and
tangential lens distortions, respectively. The shape of the error
pattern associated with the Zeiss lens looks somehow similar to
that caused by radial distortion [21]. However, the shape of the
error pattern associated with the Pentax lens looks a bit unusual
and cannot be entirely explained by radial and/or tangential
image distortions. This indicates that there is still a need for
more advanced models for correcting the different types of lens
optical distortions. In general, the results of this study are in
agreement with Refs. [21,22] in substantiating the call for carry-
ing out a calibration procedure in order to correct for the lens
distortions and thus reduce the magnitude of error in 2D-DIC
measurements.
IMAGE SHARPNESS AND CAMERA GAIN EFFECTS
ON DIC STRAIN ERROR
Fig. 8 shows a comparison of the DIC strain error values
obtained using one of the camera/lens combinations (Genie
camera with Zeiss lens) at three different imaging conditions. In
the first condition, which is the ordinary condition, the image
was well focused and the camera gain setting was set to its
default value of zero. In one of the other two conditions being
compared here, the image was slightly defocused (by changing
the focus setting of the lens), while in the other, the camera gain
was set to a high value. In the condition where the gain setting
was increased, the intensity of the illumination was reduced in
order to maintain the average image intensity level. As expected,
the results presented in the figure show that decreasing the
image sharpness reduces the DIC strain measurement accuracy.
Indeed, this observed relation between image sharpness and
strain error, confirms the conclusion drawn in the previous sec-
tion regarding the lens effect on strain error at zero translation.
The second comparison is made between the ordinary opera-
tional condition of the camera (zero gain setting) and when the
gain is set to a high value. Increasing the gain setting is an
option available in most digital cameras, and it is intended to
compensate for low illumination intensity. The figure shows
that increasing the gain increases the DIC strain error. The gain
effect seen here is rather expected since both the image intensity
levels and the random image noise are amplified when the gain
is increased.
DIRECTION OF TRANSLATION EFFECT ON DIC
STRAIN ERROR
As mentioned previously in the experimental procedure
section, the rigid-body-translation experiments performed in
this study involved translation steps in both the x- direction
and y-direction. In general, similar trends of increasing DIC
stain errors as the displacement increases can be seen for trans-
lations along the x-axis or the y-axis. Typically, researchers who
use rigid-body-translation experiments to study DIC strain
measurement accuracy perform the translations along one
direction. Moreover, all the results presented in the previous
figures are for translations along the x-axis, since it is of most
interest, because it is along the width of the image, which is
larger than the image height. In all the DIC analyses performed
in this study, the correlations were performed for a square
region of interest located at the top center of the image. By
using a square region of interest, the same number of data
points is present in both directions, and thus, the obtained
results will have the same statistical reliability in both directions.
Fig. 9 shows a comparison of the DIC strain errors, both �xx and
�yy , associated with translations along either the x-axis or the
y-axis (using the Genie camera and Zeiss lens). From the figure,
it can be seen that for zero translation, the error in both �xx and
�yy is comparable. It can also be seen that when the translation
FIG. 8 Comparison of the effects of image defocus and camera gain on DIC
strain error (Gn/Zs).
FIG. 9 Comparison of the �xx and �yy strain errors for translations in the
x-direction and y-direction (Gn/Zs).
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 13
is in the x-direction, �xx becomes clearly larger than �yy (though
it also increases), and the opposite happens when the transla-
tion is in the y-direction. Since �xx is larger when the translation
is in the x-direction, it is taken as the component that reflects
the magnitude of strain error. Similarly, �yy is taken as the com-
ponent that reflects the magnitude of strain error when the
translation is along the y-direction. Furthermore, an interesting
observation can be made from Fig. 9 regarding the magnitudes
of strain errors associated with translations in the x and y direc-
tions. By inspecting the magnitudes of the strain errors, it can
be seen that strain errors resulting from translations in the
y-direction are clearly higher than those resulting from transla-
tions in the x-direction. To further investigate this observation,
a comparison was made between the DIC strain errors associ-
ated with translations along the x-axis and the y-axis for three
different cameras, and this comparison is shown in Fig. 10. It
can be seen from the figure that the magnitude of DIC strain error
is clearly dependent on the direction of the translation where the
error is clearly higher for all the three cameras when the transla-
tion is along the y-axis. One might simply think that this differ-
ence is due to a problem in the alignment of the camera with
respect to the target; however, the fact that the same trend is seen
with the three cameras, and knowing that each one of cameras
was setup independently and later verified, makes such idea to be
unrealistic. In fact, this directionality in DIC strain errors is most
likely due to the fill factor of the imaging sensor knowing that
interline-CCD and CMOS imaging sensors have different fill fac-
tors in the x and y directions. This observation indicates that the
alignment of the image with respect to the direction of translation
will influence the magnitude of error in DIC strain measure-
ments. The results shown in Fig. 10 suggest that, during any
experiment where DIC is used to measure the strains, better accu-
racy can be obtained by aligning the camera width direction with
the direction of the maximum displacement.
Concluding Remarks
Though there is no standard approach for estimating the
errors in DIC measurements, the use of in-plane rigid-body-
translation experiments is one of the most realistic and widely
accepted methods for estimating the errors in both the displace-
ment and strain measurements of 2D-DIC analysis. In this
study, rigid-body-translation experiments were used to investi-
gate the uncertainty of DIC displacement and strain measure-
ments associated with different types of imaging systems. Four
different cameras (with different resolutions and imaging sensor
types) and four different lenses (with different optical quality
and focal length) were used in this study. By doing the rigid-
body-translation experiments using different camera/lens com-
binations, the influences of both the cameras and lenses on DIC
measurements accuracy were identified, and the magnitude of
errors associated with these different types of cameras and
lenses was determined. The influence of different imaging con-
ditions such as out-of-focus effects (image un-sharpness) and
high camera gain were also investigated. Furthermore, the influ-
ence of the direction of translation on DIC measurements accu-
racy was identified. The results of this study provide a more
thorough understanding of the contribution of the imaging sys-
tem components in the overall DIC measurements error. It is
believed that the experimental approach used in this study can
be used for quantitatively assessing the accuracy and quality of
the different types of cameras and lenses and to determine their
suitability for use in experimental techniques such as DIC or
PIV. The main conclusions of this study can be summarized in
the following points.
• In-plane rigid-body-translation, experiments are usefulfor estimating the magnitude of baseline error in 2D-DICmeasurements. Such experiments, carried under closecontrol of the experimental conditions and correlationparameters, can be used for comparing different imagingsystems and for determining the contribution of theimaging system components in the measurement error.
• Three different parameters; namely, displacement stand-ard deviation, mean of strain absolute values, and strainstandard deviation, can be used as error estimationparameters in order to determine the accuracy of DICmeasurements. The displacement standard deviation issuitable for estimating the accuracy if strain measure-ments are not required (such as the case of PIV analysis).The strain standard deviation is more suitable than themean of strain absolute values for estimating the DICstrain error, since it is more conservative.
• Reporting the displacement error in pixel units as anindependent measure of accuracy (i.e., without the scale-factor) can be misleading where it gives a false impressionthat using cameras with higher resolution will automati-cally lead to higher measurement accuracy. For numericalerror estimation studies, the displacement error can
FIG. 10 Comparison of the DIC strain errors for translations in the
x-direction and y-direction using three different cameras.
Journal of Testing and Evaluation14
simply be reported in pixels. However, for physicalstudies (i.e., ones performed using actual images), thedisplacement error “in pixels” must be reported in con-junction with the image’s scale-factor, or alternatively,the error can be reported in displacement units (e.g., mmor lm) along with the image’s magnification level.
• The estimated strain measurement error is dependent onboth the correlation and strain calculation parameters,and thus all these parameters should be carefully chosenand controlled when comparing errors associated withdifferent imaging systems. Also, it should be noted thatthe estimated strain errors are directly influenced by thechoice and values of these parameters.
• All experiments show that the measurement errorincreases as the magnitude of translation increases. Thus,when reporting the magnitude of error, it should be asso-ciated with the corresponding magnitude of rigid-body-translation.
• For the cameras tested here, the results show that thetype of camera and imaging sensor do not have a signifi-cant effect on measurement accuracy, except for colorSLR cameras, which are not designed nor meant to beused for this type of applications.
• The type and quality of the lens has a clear effect on mea-surement accuracy, and it is generally more pronouncedthan effect of the camera itself.
• The lowest DIC strain error estimate for the camera/lenscombinations used in this study is about 300 micro-strains (at 8mm translation and 7� 7 points strainwindow size), which makes the applicability of DIC foraccurately measuring small elastic strains in stiff metalslike steel to be somehow questionable especially in thecase of non-homogeneous strain fields. A calibration pro-cedure to correct for lens optical distortions will be neces-sary to improve the measurement accuracy in such cases.
• There is a clear and significant effect for the direction oftranslation on measurement accuracy. For the camerastested here, the measurement error is significantly lowerwhen the translation is along the width direction of theimage.
ACKNOWLEDGMENTS
The first author is pleased to acknowledge the financial support
provided by the Hashemite University for his sabbatical leave.
He also acknowledges the financial support provided by the
German Research Foundation (DFG) for his research visit to
the Institute of Fluid Mechanics and Aerodynamics at Universi-
tat der Bundeswehr Munchen.
References
[1] Sutton, M., Wolters, W., Peters, W., Ranson, W., andMcNeill, S., “Determination of Displacement Using anImproved Digital Correlation Method,” Image VisionComput., Vol. 1, No. 3, 1983, pp. 133–139.
[2] Sutton, M., McNeill, S., Helm, J., and Chao, Y., “AdvancesIn Two-Dimensional and Three-Dimensional ComputerVision,” Photomechanics, P. K. Rastogi, Ed., Springer, Ber-lin-Heidelberg, 2000, pp. 323–372.
[3] Sutton, M., Orteu, J., and Schreier, H., Image Correlationfor Shape, Motion and Deformation Measurements,Springer, New York, 2009.
[4] Luo, P. F., Chao, Y. J., Sutton, M. A., and Peters, W. H.,“Accurate Measurement of Three-Dimensional Deforma-tions in Deformable and Rigid Bodies Using ComputerVision,” Exp. Mech., Vol. 33, No. 2, 1993, pp. 123–132.
[5] Pan, B., Qian, K., Xie, H., and Asundi, A., “Two-Dimensional Digital Image Correlation for In-PlaneDisplacement and Strain Measurement: A Review,” Meas.Sci. Technol., Vol. 20, No. 6, 2009, 062001.
[6] Orteu, J., “3-D Computer Vision in Experimental Mechan-ics,” Opt. Lasers Eng., Vol. 47, Nos. 3–4, 2009, pp. 282–291.
[7] Willert, C. and Gharib, M., “Digital Particle Image Veloc-imetry,” Exp. Fluids, Vol. 10, No. 4, 1991, pp. 181–193.
[8] Westerweel, J., “Fundamentals of Digital Particle ImageVelocimetry,” Meas. Sci. Technol., Vol. 8, No. 12, 1997,pp. 1379–1392.
[9] Grant, I., “Particle Image Velocimetry: A Review,” Proc.Inst. Mech. Eng.: Part, C., J. Mech. Eng. Sci., Vol. 211,No. 1, 1997, pp. 55–76.
[10] Stanislas, M., Okamoto, K., Kahler, C. J., Westerweel, J.,and Scarano, F., “Main Results of the Third InternationalPIV Challenge,” Exp. Fluids, Vol. 45, No. 1, 2008,pp. 27–71.
[11] Bornert, M., Bremand, F., Doumalin, P., Dupre, J., Fazzini,M., Grediac, M., Hild, F., Mistou, S., Molimard, J., Orteu,J., Robert, L., Surrel, Y., Vacher, P., and Wattrisse, B.,“Assessment of Digital Image Correlation MeasurementErrors: Methodology and Results,” Exp. Mech., Vol. 49,No. 3, 2009, pp. 353–370.
[12] Huang, H., Dabiri, D., and Gharib, M., “On Errors of Digi-tal Particle Image Velocimetry,” Meas. Sci. Technol., Vol. 8,No. 12, 1997, pp. 1427–1440.
[13] Forliti, D., Strykowski, P., and Debatin, K., “Bias and Preci-sion Errors of Digital Particle Image Velocimetry,” Exp.Fluids, Vol. 28, No. 5, 2000, pp. 436–447.
[14] Kahler, C., Scharnowski, S., and Cierpka, C., “On theUncertainty of Digital PIV and PTV Near Walls,” Exp.Fluids, Vol. 52, No. 6, 2012, pp. 1641–1656.
[15] Chu, T., Ranson, W. F., Sutton, M. A., and Peters, W. H.,“Applications of Digital-Image-Correlation Techniques toExperimental Mechanics,” Exp. Mech., Vol. 25, No. 3,1984, pp. 232–244.
[16] Haddadi, H. and Belhabib, S., “Use of Rigid-Body Motionfor the Investigation and Estimation of the MeasurementErrors Related to Digital Image Correlation Technique,”Opt. Lasers Eng., Vol. 46, No. 2, 2008, pp. 185–196.
[17] Sutton, M., Yan, J., Tiwari, V., Schreier, H., and Orteu, J.,“The Effect of Out-of-Plane Motion on 2D and 3D DigitalImage Correlation Measurements,” Opt. Lasers Eng.,Vol. 46, No. 10, 2008, pp. 746–757.
[18] Hijazi, A., Friedl, A., and Kahler, C., “Influence of Cam-era’s Optical Axis Non-Perpendicularity on MeasurementAccuracy of Two-Dimensional Digital Image Correlation,”Jordan J. Mech. Indst. Eng., Vol. 5, No. 4, 2011,pp. 373–382.
HIJAZI AND KAHLER ON IMAGING SYSTEM EFFECT ON 2D-DIC ERROR 15
[19] Yoneyama, S., Kikuta, H., Kitagawa, A., and Kitamura, K.,“Lens Distortion Correction for Digital Image Correlationby Measuring Rigid Body Displacement,” Opt. Eng.,Vol. 45, No. 2, 2006, 023602.
[20] Zhang, D., Luo, M., and Arola, D., “Displacement/StrainMeasurements Using an Optical Microscope and DigitalImage Correlation,” Opt. Eng., Vol. 45, No. 3, 2006,033605.
[21] Pan, B., Yu, L., Wu, D., and Tang, L., “Systematic Errors InTwo-Dimensional Digital Image Correlation Due to LensDistortion,” Opt. Lasers Eng., Vol. 51, No. 2, 2013,pp. 140–147.
[22] Lava, P., Van Paepegem, W., Coppieters, S., De Baere, I.,Wang, Y., and Debruyne, D., “Impact of Lens Distortionson Strain Measurements Obtained With 2D Digital ImageCorrelation,” Opt. Lasers Eng., Vol. 51, No. 5, 2013,pp. 576–584.
[23] Reu, P., Sweatt, W., Miller, T., and Fleming, D., “CameraSystem Resolution and Its Influence on Digital ImageCorrelation,” Exp. Mech., Vol. 55, No. 1, 2015, pp. 9–25.
[24] Barranger, Y., Doumalin, P., Dupre, J. C., and Germaneau,A., “Digital Image Correlation Accuracy: Influence of Kindof Speckle and Recording Setup,” Proceedings of the ICEM14 - 14th International Conference on ExperimentalMechanics, Poitiers, France, July 4–9, 2010, F. Bremand,Ed., Vol. 6, EDP Sciences, Les Ulis, France, 31002.
[25] Tiwari, V., Sutton, M., and McNeill, S., “Assessment ofHigh Speed Imaging Systems for 2D and 3D DeformationMeasurements: Methodology Development and Vali-dation,” Exp. Mech., Vol. 47, No. 4, 2007, pp. 561–579.
[26] Hijazi, A. and Madhavan, V., “A Novel Ultra-High SpeedCamera for Digital Image Processing Applications,” Meas.Sci. Technol., Vol. 19, 2008, 085503.
[27] Burt, D., “Extending the Performance Limits of CCDImage Sensors,” GEC J. Research, Vol. 12, 1995,pp. 130–140.
[28] Hain, R., Kahler, C., and Tropea, C., “Comparison of CCD,CMOS and Intensified Cameras,” Exp. Fluids, Vol. 42,No. 3, 2007, pp. 403–411.
[29] MatchID, 2013, “MatchID-2D Manual,” www.MatchID.org,(Last accessed on 28 June 2013)
[30] Patterson, E., Hack, E., Brailly, P., Burguete, R. L., Saleem,Q., Siebert, T., Tomlinson, R. A., and Whelan, M. P.,“Calibration and Evaluation of Optical Systems for Full-Field Strain Measurement,” Opt. Lasers Eng., Vol. 45,No. 5, 2007, pp. 550–564.
[31] Hoult, N., Andy Take, W., Lee, C., and Dutton, M.,“Experimental Accuracy of Two Dimensional Strain Meas-urements Using Digital Image Correlation,” Eng. Struct.,Vol. 46, 2013, pp. 718–726.
[32] Siddiqui, M. and Ahmed, M., “An Out-of-Plane MotionCompensation Strategy for Improving Material ParameterEstimation Accuracy With 2D Field Measurements,”Exp. Mech., Vol. 54, No. 7, 2014, pp. 1259–1268.
[33] Hild, F. and Roux, S., “Digital Image Correlation: FromDisplacement Measurement to Identification ofElastic Properties–A Review,” Strain, Vol. 42, No. 2, 2006,pp. 69–80.
[34] Rossi, M., Pierron, F., and Forquin, P. A., “Assessment ofthe Metrological Performance of an In Situ Storage ImageSensor Ultra-High Speed Camera for Full-Field Deforma-tion Measurements,” Meas. Sci. Technol., Vol. 25, No. 2,2014, 025401.
[35] Dereniak, E. and Dereniak, T., Geometric and Trigonomet-ric Optics, Cambridge University Press, Cambridge, UK,2008.
Journal of Testing and Evaluation16
Copyright by ASTM Int’l (all rights reserved); Thu Mar 17 14:10:45 EDT 2016Downloaded/printed byAla Hijazi (The Hashemite University, Mechanical Engineering Dept., P.O. Box 330127, Zarqa, Jordan, 13133)Pursuant to License Agreement. No further reproduction authorized.